Skip to main content

Module operator_runtime

Module operator_runtime 

Source
Expand description

Compatibility forwarding to the canonical Etalon owner.

Modules§

adapter
Prepared operator adapter binding representation digest and residency generation (W3: EOS-031).
cpu_lookup
Optimized CPU lookup execution kernel (W6: EOS-060).
gpu_schedule
Forge GPU decode/prefill dual-path schedules and occupancy diagnostics (W6: EOS-061).
moe_operator
Generic independent and clustered expert execution.

Structs§

ClusterAnchor
A cluster anchor containing shared base weights and per-expert deltas (EOS-071).
ClusteredMoEOperator
Clustered MoE operator holding cluster anchors with shared base projections and per-expert deltas (EOS-071).
ClusteredMoeScratch
Scratch buffers required for zero-heap clustered MoE evaluation.
CpuScheduleConfig
Configuration for CPU lookup execution.
GpuOccupancyDiagnostics
Analytical occupancy diagnostics for a synthesized Forge schedule.
GpuOperatorSchedule
Synthesizes a GPU execution schedule descriptor for an operator matrix multiplication.
IndependentExpertWeights
Independent SwiGLU expert weights (EOS-070).
LowRankDelta
A low-rank residual factor delta: delta_W * x = A * (B * x).
PreparedOperator
An operator prepared for execution within a residency generation.
RoutedExpert
Routed expert execution item: (expert_index, gating_weight).

Enums§

GpuScheduleKind
Target execution path on the GPU.
PreparedOperatorError
Errors occurring during operator preparation or execution.

Functions§

dispatch_clustered_moe_step
Execute clustered MoE dispatch step with down-projection pre-accumulation (EOS-071).
execute_cpu_q4k_lookup
Execute CPU lookup GEMV for GGML Q4_K weights against an input vector.
precompute_activation_luts
Precomputes 16-entry nibble lookup tables for each 32-element subblock in an input vector.